Our general approach for compression comparisons is outlined by the following block diagram:
The compressed image similarity with the original is plotted as a function of the compression ratio for each of the methods we implemented below. There were also several points for SVD that had much higher compression ratios than the plot below, but were left off for legibility reasons. SVD has a much higher possible compression ratio, since spatial information doesn’t need to be kept using this method, but it needs to be represented with single bits using the other methods. Huffman encoding isn’t included in this plot since it is lossless.
For lower compression ratios, Haar Wavelet compression had higher image similarities than SVD and DCT Compression as we have implemented them. For any compression ratio above 8X, however, SVD was by far superior, since that algorithm was able to achieve much higher compression ratios but at considerable image quality loss. Compression ratios of 63.94X and 255.75X were achieved, but the images were only 78% and 69% similar to the original image respectively.
Image Compression 